Cognition

Latest AI news, models and releases from Cognition. ['Devin', 'FrontierCode', 'SWE-1.7']

AI Safety 🇺🇸

Import AI 461: 'Alignment is not on track'; FrontierCode; and synthetic research interns

Researchers from the UK AI Security Institute and Timaeus launch Sequent, a nonprofit aiming to develop principled alignment techniques for superintelligent AI, raising $100-150M initially. Cognition releases FrontierCode, a hard coding benchmark where Claude Opus 4.8 scores 13.4% on the hardest tier. Xiaomi publishes details on MiMo-V2.5-Pro-UltraSpeed, a 1-trillion parameter model achieving 1000 tokens per second via co-design and quantization. A new benchmark, AARRI-Bench, evaluates AI systems on entry-level research tasks.

CognitionCognition Alibaba/QwenAlibaba/Qwen AnthropicAnthropic OpenAIOpenAI
Import AI27.07 · 21:04
Regulation 🇨🇳

Nvidia, Microsoft, IBM and Dozens of Other Companies Form AI Safety Alliance

Nvidia, Microsoft, IBM, Adobe, SAP and other companies have announced the formation of the Open Alliance for AI Safety (OASIS). The alliance will focus on developing and distributing open tools for the safe and responsible use of AI.

NVIDIANVIDIA MicrosoftMicrosoft IBMIBM SAPSAP DellDell Cisco SystemsCisco Systems Cadence Design SystemsCadence Design Systems Capital OneCapital One ClouderaCloudera CloudflareCloudflare CognitionCognition CrowdStrikeCrowdStrike DatabricksDatabricks DoorDashDoorDash ElasticElastic Hugging FaceHugging Face LangChainLangChain Linux FoundationLinux Foundation NetAppNetApp Nous ResearchNous Research PalantirPalantir Palo Alto NetworksPalo Alto Networks Red HatRed Hat SalesforceSalesforce SK TelecomSK Telecom SiemensSiemens SnowflakeSnowflake Synopsys, Inc.Synopsys, Inc.
36Kr27.07 · 13:03
Fresh news